How can I maximize my PPD?

[DR650SE]

Below are the three systems I have. All running windows 7 and the V7 beta of F@h. All systems are running SMP and GPU/GPUs. For 7 total clients.

Dell Inspiron 1720 Laptop. 2.6GHz Core 2 Duo CPU. 8600M GT GPU

Alienware M11x Laptop. 1.6GHz Core 2 Duo CPU. 335M GPU

Alienware M17x R2 Laptop. i7 940XM 2.5GHz Quad core. Runs 2.5GHz folding full load. Dual AMD 6990M GPUs.

I know I can OC the components to boost production, but I don't want to have to test stability and what not. Mostly because I go out of town a lot. So I'd rather have peace of mind knowing the systems are folding with no issues while I'm gone.

What can I do to maximize my PPD? Should I drop everything to small work units? Should I run large work units? Should I only run SMP on all systems and not bother with GPU? Should I only run GPUs? If I run GPUs and CPU on the same system, is it actually pull down my PPD because now the CPU is folding as well as running other tasks? Should I disable hyperthreading on the i-7 CPU?

One thing I do is monitor my computers while I'm out of town. I use Team viewer to remote into my 3 notebooks from my Droid phone so I can make changes to the systems, even if I'm several states away, which I normally am. And advice is greatly appreciated. Thanks guys.

Bottom line: What settings should I use to maximize PPD?

[Zagen30]

For the Inspiron, SMP+GPU is probably the best option. I don't think the GPU will earn that much, but the pre-Fermi architectures use essentially zero CPU time, so there shouldn't be any noticeable SMP PPD hit. If there is, though, the GPU would likely be the first thing to go.

The M11x is in the same boat; the 335 is pre-Fermi, so it shouldn't affect the SMP performance. This machine has the highest potential of not meeting SMP deadlines, so if that's the case you may want to switch to 2xuniprocessor and 1xGPU.

The M17x is probably the one you'll have to experiment with the most. AMD cards need a lot of CPU time due to the way the drivers are written. You're definitely not going to get max performance running SMP:8 and 2xGPU. I'm not overly familiar with how well the multithreading of an i7 can handle AMD GPU needs, but as a starting point I'd try SMP:6 alongside both GPUs; make sure you put in "client-type - advanced" to get the more efficient Core 16 GPU WUs if you're not already doing so. Disabling HT would probably hurt things, since the multithreading can be rather helpful. You'll probably want to see if SMP:5 produces any noticeably better results. SMP:7 likely wouldn't offer up enough CPU time, and certain projects can't handle large prime numbers of threads; I don't believe 5 is large enough to cause problems, but I may be wrong, in which case you'd want to check SMP:4.

As to WU size, I'm not sure it'll make that much of a difference. I believe all SMP projects are large anyway (regardless of what option you give), and there's only a few of the Core 16 projects right now, so I don't believe it'd affect the AMD GPUs either. You might want to see if the small WU selection on the two Nvidia cards is any better (or different at all) from larger projects since they're somewhat low in shader count. If you end up running uniprocessor slots on any of the machines it might make more of a difference, but I haven't folded uniprocessor in years so I can't really comment on that.

[DR650SE]

Thanks, I will defintely be doing some experimenting with the M17x.

[weirddan456]

I have a i5 2500K quad core and a Radeon 5970, which is a dual GPU card. What I had to do is devote a full core per GPU slot, so that means 2 GPU slots and SMP running on the other 2 cores. I don't know how hyperthreading will work with that though, my i5 doesn't support it. If the SMP doesn't get full access (or nearly full anyway) to all the cores you allocate it to, PPD drops dramatically.

[bruce]

I have a i5 2500K quad core and a Radeon 5970, which is a dual GPU card.... I don't know how hyperthreading will work with that though.... If the SMP doesn't get full access (or nearly full anyway) to all the cores you allocate it to, PPD drops dramatically.

If you replace your i5 with an i7, you'll still need two cores for the GPUs. You'll be able to run either [SMP6 plus two GPUs] or [SMP4 plus two GPUs plus two free virtual cores]. There are still only four FPUs shared across the 8 virtual cores, so SMP6 and SMP4 will be both quite similar to what you would be getting on the i5 if you stopped the GPUs and ran SMP4. In this special case, adding two GPUs uses the "free" resources created by HT.

Same question being discussed here: viewtopic.php?f=15&t=20543